Interaction of tumor cells and astrocytes promotes breast cancer brain metastases through TGF-β2/ANGPTL4 axes

Metastatic outcomes depend on the interactions of metastatic cells with a specific organ microenvironment. Our previous studies have shown that triple-negative breast cancer (TNBC) MDA-MB-231 cells passaged in astrocyte-conditioned medium (ACM) show proclivity to form brain metastases, but the underlying mechanism is unknown. The combination of microarray analysis, qPCR, and ELISA assay were carried out to demonstrate the ACM-induced expression of angiopoietin-like 4 (ANGPTL4) in TNBC cells. A stable ANGPTL4-knockdown MDA-MB-231 cell line was generated by ANGPTL4 short-hairpin RNA (shRNA) and inoculated into mice via left ventricular injection to evaluate the role of ANGPTL4 in brain metastasis formation. The approaches of siRNA, neutralizing antibodies, inhibitors, and immunoprecipitation were used to demonstrate the involved signaling molecules. We first found that ACM-conditioned TNBC cells upregulated the expression of ANGPTL4, a secreted glycoprotein whose effect on tumor progression is known to be tumor microenvironment- and tumor-type dependent. Knockdown of ANGPTL4 in TNBC MDA-MB-231 cells with shRNA decreased ACM-induced tumor cell metastatic growth in the brain and attributed to survival in a mouse model. Furthermore, we identified that astrocytes produced transforming growth factor-beta 2 (TGF-β2), which in part is responsible for upregulation of ANGPTL4 expression in TNBC through induction of SMAD signaling. Moreover, we identified that tumor cells communicate with astrocytes, where tumor cell-derived interleukin-1 beta (IL-1β) and tumor necrosis factor alpha (TNF-α) increased the expression of TGF-β2 in astrocytes. Collectively, these findings indicate that the invading TNBC cells interact with astrocytes in the brain microenvironment that facilitates brain metastases of TNBC cells through a TGF-β2/ANGPTL4 axis. This provides groundwork to target ANGPTL4 as a treatment for breast cancer brain metastases.


INTRODUCTION
The incidence of breast cancer brain metastasis (BM) has increased in recent years, particularly in the subpopulation of triple-negative breast cancer (TNBC) (estrogen receptor <ER>-/progesterone receptor <PR>-/human epidermal growth factor receptor 2 <Her2>-, ER-/PR-/Her2-). 1-3 TNBC, which accounts for 15-25% of all breast cancers, 4,5 has problematic diagnosis because patients with TNBC do not respond to hormone or HER2-targeted therapies due to the lack of expression of ER, PR, and HER2. 4 Moreover, the exact pathologic process of BM in the TNBC subset is poorly understood up to now. These facts emphasize the clinical importance of BM in TNBC and provide an explanation for why patients with TNBC have the worst outcomes out of all breast cancer patients. Therefore, much effort is needed on this front to help address BM in TNBC.
Ultimately, the clinical outcomes of people with metastatic cancer greatly rely on the interactions of the metastatic tumor cells and the cells in the surrounding microenvironment of an organ. 6,7 Astrocytes, a subtype of glial cells, are the most abundant cell type in the brain microenvironment. In breast cancer BM patients, reactive astrocytes accumulate around the metastatic foci; 8-10 they confine brain metastases without infiltrating the lesion. 11 Previous studies have found that reactive astrocytes participate in tumor progression and chemoresistance through direct physical contact and gap junctional communication with tumor cells in the brain. 12,13 It has also been reported that paracrine signaling via cytokines, growth factors, or exosomes released by astrocytes facilitates tumor metastasis formation in the brain and affects blood-brain barrier permeability. 9,10,14,15 On the other hand, the interaction between tumor cells and astrocytes may not be just synergistic. There was a report that brain-metastatic cells express high-level serpins to prevent astrocyte-generated plasmin and its metastasis-suppressive effects. 16 Although these studies have provided strong evidence about the role of astrocyte tumor BM, the underlying mechanisms remain poorly understood.
We have previously shown that astrocyte-conditioned medium (ACM) facilitated metastasis formation of MDA-MB-231 cells in the brain. 17 In this follow-up study, we further investigated the underlying mechanism for the ACM-induced BM phenotype. After conditioning TNBC cells in ACM, we found that ACM upregulated the expression of angiopoietin-like 4 (ANGPTL4) in TNBC cells. ANGPTL4 is a secreted glycoprotein whose effect on tumor progression is tumor-type-and microenvironment-dependent. [18][19][20] It has been shown that transforming growth factor-beta 1 (TGF-β1) induced ANGPTL4 expression in breast cancer cells, which facilitated breast cancer cells to metastasize to the lung. 21 Mouse astrocytes secrete TGF-β2, 22 a homolog protein of TGF-β1. 23,24 On this basis, we investigated how tumor cells and astrocytes interact to contribute to breast cancer BM through the TGF-β2/ ANGPTL4 axis.

ACM upregulates ANGPTL4 expression in TNBC cells
The most intriguing finding from our earlier studies is that after passaging the MDA-MB-231 cells in ACM for five passages, the invasion competence of the tumor cell showed a sudden and rapid increase, which led to an increase in vitro invasion when compared with the cells passaged 1-4 times in ACM (fold change to cells passaged five times in DMEM: 1.9, 2.2, 6.8, 7.4, and 19.8). 17 During inoculation of the passage 5 cells into mice through left ventricular injection, rapid and severe brain metastases were developed. 17 To gain insight into the mechanism of the BM, we investigated whether the ACM affected gene expression by microarray expression analysis using MDA-MB-231/P5A, MDA-MB-231/P5D (control), and MDA-MB-231/P10AD (determining if the ACM elicited permanent or transient effects on the cells). Based on the data of the 15,453 gene-level probe sets, we sought for the genes with significant expression changes (p < 0.01 and FC > 2.0) in response to ACM. The distribution of genes according to the significance of their expression was represented in volcano plots (Fig. 1a). We also generated a hierarchical clustering heatmap across the nine samples under the three different cultures (Fig. 1b). Our results showed that the expression of 3676 genes was significantly changed with 1875 upregulated genes and 1801 downregulated genes in MDA-MB-231/P5A vs MDA-MB-231/P5D (Fig. 1a upper panel. Additional file 2 Table S1 for Table S2 for gene lists). Of interest, clustering analyses across the nine samples under the three cultures grouped MDA-MB-231/P10AD and MDA-MB-231/P5D in the same cluster ( Fig.  1b), which hints that the expression genes from MDA-MB-231/ P10AD cells are more similarly relative to the genes from MDA-MB-231/P5D cells. Together, these results suggest that ACM may induce and maintain genetic alteration in MDA-MB-231 cells.
Based on our microarray analysis, ANGPTL4 is one of the most upregulated genes. ANGPTL4 has been recently emerging as an important factor in tumor progression. 18,21,25 Therefore, qPCR was performed to confirm the expression of ANGPTL4 in the three cells (Fig. 1c). ANGPTL4 expression was significantly higher in MDA-MB-231/P5A cells when compared with MDA-MB-231/P5D (p = 0.010 by the two-sided Student's t-test) and MDA-MB-231/P10AD cells (p = 0.018 by the two-sided Student's t-test), which suggests that ACM induced and maintained the expression of ANGPTL4 in MDA-MB-231 cells.
To examine whether ACM upregulates ANGPTL4 expression in other TNBC cells, TNBC MDA-MB-231, MDA-MB-468, HCC1937 cells, breast cancer estrogen receptor-positive cells (MCF-7), and immortalized breast epithelial cells (MCF-10A) were sequentially passaged in ACM for five passages. Cells passaged in the corresponding cell culture media (CM) were used as control. Gene expression was then analyzed by qPCR. Our data showed that basal expression of ANGPTL4 in cells cultured in media (CM) was similar (Additional file 3: Fig. S1a). However, after being passaged in ACM, the expression of ANGPTL4 significantly increased in all tumor cells compared with CM control (p < 0.05 in MCF-7 and p < 0.005 in TNBC cells by the two-sided Student's ttest, Fig. 1d). Moreover, the expression of ANGPTL4 significantly increased in all TNBC cells compared with MCF-10A cells (p < 0.01 by the two-sided Student's t-test, Fig. 1d), as well as in MDA-MB-231 and MDA-MB-468 cells compared with MCF-7 cells (p < 0.005 by the two-sided Student's t-test, Fig. 1d). Furthermore, we examined protein expression of ANGPTL4 in cell lysates and culture supernatants by ELISA (Fig. 1e). We found that the ANGPTL4 protein expression significantly increased in all tumor cells when compared with CM control (p < 0.05 by the two-sided Student's t-test), and MDA-MB-231 and HCC1937 cells in ACM expressed significantly higher ANGPTL4 than MCF-10A and MCF-7  Fig. S2a), these cells were injected into 6-8-week-old male CB17/SCID mice through the left ventricle, and bioluminescence imaging was performed to visualize brain-metastatic lesions (Additional file 3: Fig. S2b, c and Fig. 2c). Brain metastases were evaluated. Total flux was used to monitor tumor growth. We found that the total brain flux at day 21 was significantly lower in A4shRNA/ACM compared with ConshRNA/ACM (p = 0.029 by the two-sided Student's t-test, Fig.  2d). On the last day of measurement (either the day of death or day 42 for those mice that survived), total brain flux had no statistically significant difference between the two groups (p = 0.31 by the two-sided Student's t-test. Additional file 3: Fig.  S2d). When comparing the time to BM between the two groups, we also did not see significant differences (p = 0.36 by the logrank test. Additional file 3: Fig. S2e). However, if we compared the fold change of ConshRNA/ACM with respect to A4shRNA/ACM on the last day, total flux was more than two fold, and BM incidence was about 1.4-fold (Additional file 3: Fig. S2c, e). Moreover, the survival of mice was about 3.14-fold compared with MB-231/ A4shRNA and MDA-MB-231/ConshRNA cells (4 out of 7 and 2 out of 11. Additional file 3: Fig. S2c). Together, these results suggest that knockdown of ANGPTL4 decreases MDA-MB-231 cell metastatic growth in the brain.
TGF-β2 derived from astrocytes upregulates the expression of ANGPTL4 in TNBC cells TGF-β1 has been shown to induce ANGPTL4 expression in TNBC cells, leading to pulmonary metastases. 21 There are reports that TGF-β2, but not TGF-β1, is present in the culture supernatants of mouse astrocytes, 22 and cultured human adult astrocytes produce biologically active TGF-β2 but only minimal TGF-β1. 26 Because TGF-β2 and TGF-β1 are highly homologous, 23,24 we investigated whether astrocyte-secreted TGF-β2 can regulate ANGPTL4 expression. TGF-β2 was detected in both ACM and hACM by ELISA (Fig. 3a). We then sought to examine whether astrocyte-secreted TGF-β2 induces ANGPTL4 expression in TNBC cells. To do so, we first examined whether the exogenous addition of TGF-β2 could induce ANGPTL4 expression in TNBC cells. MDA-MB-231 and MDA-MB-468 cells treated with human recombinant TGF-β2 (5 ng/ml, Pepro Tech, Inc.) showed that TGF-β2 upregulated both RNA and protein expression of ANGPTL4 (p < 0.01 by the two-sided Student's t-test, Fig. 3b). We then analyzed whether astrocyte-secreted TGF-β2 was Fig. 1 ACM upregulates ANGPTL4 expression in TNBC cells. a Volcano plots for differential ACM-responsive genes. The log-fold change relative to P5D is plotted on the x-axis (the downregulated genes represented by green points, the upregulated genes represented by red points) and the negative log10 p-value is plotted on the y-axis. The black, dashed line represents the p-value cutoff (0.01, Student's t-test) and log FC cutoff (±1.0). b Heatmap of differential ACM-responsive gene expression. Heatmap of gene expression alterations seen across all probe sets, with triplicate samples per culture condition. Each column represents a single culture and each row represents a single gene. Red indicates overexpressed probe sets, whereas green indicates underexpressed probe sets. The top cluster diagram indicates the hierarchical clustering order of the individual samples analyzed. The right cluster diagram represents differentially expressed probes with p < 0.01 and FC > 2.0. As shown, ANGPTL4 is one of the top overexpressed genes. c qPCR of ANGPTL4 transcripts in ACM-conditioned MDA-MB-231 cells. *p < 0.05 compared with P5D and + p < 0.05 compared with P10AD by the two-sided Student's t-test, n = 3. d qPCR of ANGPTL4 transcripts in ACM-conditioned TNBC cells. *p < 0.05 compared with CM control, + p < 0.05 compared with MCF-10A in ACM, and # p < 0.05 compared with MCF-7 in ACM by the two-sided Student's t-test. n = 3. e ELISA analysis of the expression of ANGPTL4 in supernatants and cell pellets of the ACM-conditioned TNBC cells. *p < 0.05 compared with CM control, + p < 0.05 compared with MCF-10A in ACM, and # p < 0.05 compared with MCF-7 in ACM by the two-sided Student's t-test. n = 3. f ELISA analysis of ANGPTL4 in hACMconditioned TNBC cells. *p < 0.05 compared with CM control, + p < 0.05 compared with MCF-10A in hACM, and # p < 0.05 compared with MCF-7 in hACM by the two-sided Student's t-test. n = 3. From (c) to (f), CM: tumor cell culture medium; ACM: rat astrocyte-conditioned medium; hACM: human astrocyte-conditioned medium c Representative brain metastasis tumor images. d Total brain flux at 21 days post injection. *p = 0.029 by the two-sided Student's t-test comparing A4shRNA/ACM (mean 2.51e + 6, SD 2.56e + 6) with ConshRNA/ACM (mean 5.35e + 7, SD 1.29e + 8) involved in the induction of ANGPTL4 in TNBC cells. TGF-β2depleted ACM was used to culture the TNBC cells for five passages. A reduction of ANGPTL4 in supernatants and cell lysates from cells cultured in TGF-β2-depleted ACM was observed when compared with cells cultured in regular ACM (p < 0.05 by the two-sided Student's t-test, Fig. 3c). Collectively, these data suggest that astrocytes secrete TGF-β2, which upregulates the expression of ANGPTL4 in TNBC cells.
TGF-β2 upregulates the expression of ANGPTL4 in TNBC cells through canonical TGF-β pathway On the above basis, the upstream signaling pathway involved in the induction of ANGPTL4 by TGF-β2 in TNBC cells was investigated. It is well known that TGF-β signals via its transmembrane serine/threonine kinase receptors. Binding of TGF-β to TGF-βR2 recruits TGF-βR1 into the receptor complex, resulting in the phosphorylation of TGF-βR1 to transmit the signal to downstream signaling molecules. 27,28 To examine whether TGF-βR is involved in the ANGPTL4 induction, MDA-MB-231 and MDA-MB-468 cells were first treated with TGF-βR1 inhibitors, SB431542 and SB525334 (Selleck Chemicals), respectively, and then stimulated with TGF-β2 (5 ng/ml). The results showed that blocking TGF-βR1 function reduced TGF-β2-induced ANGPTL4 expression in TNBC cells (p < 0.05 by the two-sided Student's t-test, Fig. 4a). These data suggest that TGF-βR1 mediates the induction of ANGPTL4 by TGF-β2.
Phospho-TGF-βR1 activates transcription factors SMAD2 and SMAD3 via phosphorylation. Phosphorylated SMAD2/3 subsequently forms a complex with co-SMAD (SMAD4) and enters the nucleus to regulate gene transcription. 27 Thus, SMAD participation in the induction of ANGPTL4 by TGF-β2 was examined. After treating MDA-MB-231 cells with TGF-β2 (5 ng/ml) for 6, 24, and 48 h, the expression of phospho-SMAD3 was increased (Fig. 4b shows the expression of ANGPTL4 at 6 h). The SMAD3/SMAD4 complex formation was induced after 24-h treatment with TGF-β2 in MDA-MB-231 cells (Fig. 4c). The involvement of SMAD3/ SMAD4 signaling in TGF-β2-induced ANGPTL4 expression was further verified when knockdown of SMAD3/SMAD4 reduced ANGPTL4 expression in MDA-MB-231 cells (p < 0.05 by the twosided Student's t-test, Fig. 4d). Collectively, these data suggest that the canonical TGF-β/SMAD pathway mediates TGF-β2-induced ANGPTL4 expression in TNBC cells. TNBC cell-derived IL-1β and TNF-α enhance the expression of TGF-β2 in astrocytes Astrocytes are typically in an inactive, quiescent state in vivo. However, in response to brain injury, stroke, or brain cancer pathologic conditions, astrocytes become activated, a process named reactive gliosis, which is associated with profound changes in gene expression. [29][30][31][32] Although astrocytes also tend to be very reactive in cell culture, 33 astrocyte activity is a highly heterogeneous state depending on the specific injury. 34 Thus, whether TNBC cells affect the TGF-β2 expression by astrocytes was examined. Following the demonstration of the low TGF-β2 expression level in tumor-conditioned medium (TCM) from the cultures of TNBC MDA-MB-231, MDA-MB-468, and HCC1937 cells in serum-free breast cancer cell line media (Fig. 5a left panel), the media were used in culture with astrocytes, respectively. Astrocytes cultured in the corresponding serum-free tumor cell culture medium (SFM) that was not preconditioned by tumor cells was used as control. After 24 h in the TCM, the amount of TGF-β2 produced by the astrocytes was measured by ELISA. A significant increase in TGF-β2 in the resultant astrocyte culture media was observed compared with astrocytes in TCM with those in SFM (p < 0.005 in MCF-10A, p = 0.0001 in MCF-7, and p < 0.0001 in TNBC cells by the two-sided Student's t-test, Fig. 5a right panel), as well as compared with astrocytes in TCM from TNBC cells with those in non-TNBC cells (p < 0.0005 in MCF-10A and p < 0.05 in MCF-7 by the two-sided Student's t-test, Fig. 5a

right panel). This suggests that breast tumor cells, especially TNBC cells secrete factors into the TCM that increase the expression of TGF-β2 in astrocytes.
It has been reported that IL-1β and TNF-α increased the expression of TGF-β2 in human retinal pigment epithelial cells. 35 Both IL-1β and TNF-α are highly expressed in invasive ductal carcinoma (IDC). 36 Also, 90% of TNBCs are considered to be IDC. 37 Therefore, TNBC cells may induce TGF-β2 expression in astrocytes through IL-1β and TNF-α produced by TNBC cells. To test this possibility, we first accessed transcriptome data of different breast cancer subtypes from TCGA. By analyzing the data from 719 breast cancer samples in which the luminal is 566, Her2 + is 37, and TNBC is 116, we found that the expression of IL-1β and TNF-α showed a significant increase in TNBC compared with luminal and Her2 + breast cancers (p < 0.005 by the two-sided Student's t-test, Fig. 5b). Then, the significant higher expression level of IL-1β and TNF-α from TNBC cells compared with non-TNBC cells was confirmed by ELISA (p < 0.005 for IL-1β and p < 0.05 for TNF-α by the two-sided Student's t-test, Fig. 5c). After that, TNF-α and IL-1β (separately and together) within the TCM were first neutralized by using neutralizing antibodies, and the resultant TCMs were then applied to astrocyte cultures. The results showed that neutralization of the two cytokines, both separately and together, led to decreased TGF-β2 expression by astrocytes (p < 0.05 by the twosided Student's t-test, Fig. 5d). In addition, after treating astrocytes with human recombinant IL-1β (10 ng/ml, R&D Systems) and TNFα (10 ng/ml, R&D Systems) for 24 h in serum-free media, the resultant culture supernatant revealed higher expression levels of TGF-β2 than the culture supernatant from non-treated astrocytes (p < 0.005 by the two-sided Student's t-test, Fig. 5e). Together, these results suggest that TNBC cell-secreted IL-1β and TNF-α increase the expression of TGF-β2 in astrocytes.

DISCUSSION
The development of BM occurs through multiple steps. Colonization and growth of metastatic tumor cells in the brain parenchyma is the most complex and rate-limiting phase of metastasis, which is dependent on the cross-talk between tumor cells and the brain microenvironment. 38,39 Studies from human breast cancer-brain Astrocytes cultured in serum-free breast cancer cell line media (SFM) that was not preconditioned by tumor cells were used as control. *p < 0.05 compared with astrocytes in TCM with SFM; + p < 0.05 compared with MCF-10A in TCM and # p < 0.05 compared with MCF-7 in TCM by the two-sided Student's t-test. n = 3. b Mean centralized the mRNA of IL-1β and TNF-α in three subtypes of breast cancer. Data from TCGA 719 breast cancer cohort. ***p < 0.005 compared with luminal and Her2 + breast cancers by the two-sided Student's t-test. c TNBC cells produced IL-1β and TNF-α. TCM from different cells was analyzed by ELISA to quantity the levels of IL-1β (left) and TNF-α (right), respectively. *p < 0.05 compared with MCF-10A in TCM and + p < 0.05 compared with MCF-7 in TCM by the twosided Student ' s t-test. n = 3. d TNBC cell-derived IL-1β and TNF-α enhanced TGF-β2 expression in astrocytes. TCM treated with or without neutralizing anti-IL-1β or/and anti-TNF-α antibodies was used to culture astrocytes, and the resultant media were analyzed by ELISA to quantity TGF-β2 expression. SFM: serum-free medium. TCM: tumor cell-conditioned medium. # p < 0.01 compared with SFM, *p < 0.05, **p < 0.01, and ***p < 0.005 compared with TCM by the two-sided Student's t-test. n = 3. e IL-1β and TNF-α induced TGF-β2 expression in astrocytes. Astrocytes treated with vehicle or IL-1β (10 ng/ml, 6 h) or/ and TNF-α (10 ng/ml, 6 h) were analyzed by ELISA to quantify TGF-β2 expression. *p < 0.05, **p < 0.01, and ***p < 0.005 compared with vehicle by the two-sided Student's t-test. n = 3 xenograft mouse models and clinical cases found that brainmetastatic colonization is accompanied by an extensive response in the surrounding brain tissue, recruiting large numbers of reactive astrocytes, which further associate with metastatic brain lesions. 8,9,11 Our published data have shown that the ACMpassaged MDA-MB-231 cells demonstrated a higher incidence of tumor cell seeding in the brain and developing brain lesions in a mouse model compared with MDA-MB-231 cells passaged in DMEM. 17 A recent study also reported that ACM induced cell elongation and promoted actin stress fiber organization in metastatic breast cancer cells. 40 These observations suggest that astrocytes may derive cytokine and growth factors through paracrine signaling to promote the ability of breast cancer cells to seed and grow in the brain. In this study, we identified that astrocytes secrete TGF-β2 to upregulate ANGPTL4 expression in TNBC cells, which contributed to tumor cell brain lesions.
ANGPTL4 is a secreted protein belonging to a family of angiopoietin-like proteins, which has been known for its role as an adipokine involved in the regulation of lipid and glucose metabolism. 41 It has been reported that upregulation of ANGPTL4 was observed in various types of human cancers, such as colorectal cancer and breast cancer. 18,21,25 Elevated ANGPTL4 expression in these cancers correlates to enhanced metastasis, tumorigenesis, and a poor prognosis, while downregulation of ANGPTL4 leads to the opposite effect. 18,19,21,25 However, studies have also shown that increased ANGPTL4 expression inhibits melanoma, lung, and colorectal tumor growth, metastasis, and angiogenesis. 20,42 Thus, the expression and the role of ANGPTL4 in tumors appears to be context-and tumor-type dependent, which suggests the clinical importance of ANGPTL4 in tumor biology. As for breast cancer BM, a previous study has revealed that ANGPTL4 is one of 17 genes within the breast cancer brain metastasis gene set (BrMS) whose expression was correlated with brain relapse in clinically annotated breast tumors and resembled the expression profile of brain-metastatic-derived cells from a mouse model. 43 In this study, we found that knockdown of ANGPTL4 in MDA-MB-231 cells significantly reduced the ability of these tumor cells to seed and grow in the brain at 21 days post injection. The significance at later time points may have been slightly skewed by the fact that murine Angptl4 is highly homologous to human ANGPTL4. Therefore, Angptl4 produced from other, non-tumor cell tissues in the mouse, may cause more variation in the results. For example, adipocyte-derived ANGPTL4 drives disease progression under obese conditions, thus demonstrating that ANGPTL4 produced from other cell types may still promote tumor progression. 44 However, this finding is a direct evidence for the tumor-promoting role of ANGPTL4 in breast cancer BM, which provides groundwork to warrant further investigation into targeting ANGPTL4 as a treatment for breast cancer BM.
By focusing on TGF-β2, a homolog of TGF-β1, this study identified that the induction of ANGPTL4 in TNBC cells is through astrocyte-derived TGF-β2 (Fig. 3). However, TGF-β2 is only partially responsible for the ANGPTL4 induction by ACM, as tumor cell ANGPTL4 induction in the full ACM was about 1.6-fold compared with that in the TGF-β2-depleted ACM, whereas it was about threefold compared with that in DMEM. Thus, the ANGPTL4 expression from the cells passaged in the TGF-β2-depleted ACM is almost twofold of the cells passaged in DMEM. This indicates that TGF-β2 is just responsible for about 50% of ACM-induced ANGPTL4 upregulation. Considering the hundreds of cytokines and growth factors in ACM, this 50% of effect makes TGF-β2 one of the major factors derived by astrocytes that regulate the expression of ANGPTL4 in brain-metastatic breast cancer cells. Furthermore, this study identified that TGF-β2 induces ANGPTL4 expression through TGF-βR to activate Smad3, which leads to the formation of Smad3/Smad4 complex to upregulate the expression of ANGPTL4. SMAD4 knockdown reduced TGF-β2-mediated induction of ANGPTL4 expression, suggesting its involvement in metastasis. While SMAD4 is generally considered a tumorsuppressor protein, it has also been demonstrated to be involved in metastasis of breast cancer. 45 Specifically, SMAD4 knockdown in MDA-MB-231 cells significantly inhibited bone metastasis when injected into nude mice. 45 In agreement with this study, our findings suggest that SMAD4 involvement in tumor progression may depend on a variety of factors including tumor microenvironment. Another important fact that we observed in this study is that the upregulated ANGPTL4 level by ACM was decreased after re-passaging the cells in the DMEM, which shows that the regulation of ANGPTL4 is astrocyte dependent and is reversible.
In response to specific pathological conditions, astrocytes become activated by differently changing their gene expression in the process of reactive gliosis. In brain-metastatic melanoma, astrocytes were able to activate and produce more heparanase to promote the brain colonization of melanoma cells. 46 In response to HER2-transfected MDA-MB-231 breast cancer cells, astrocytes activated and expressed phosphorylated p751-PDGFRβ, which in turn contributed to the brain-metastatic growth of the tumor cells. 47 Brain-metastatic cells induce and maintain STAT3 activity in reactive astrocytes surrounding metastatic lesions that are required for BM. 11 This astrocyte characteristic challenges our current approach of using ACM collected from in vitro cell culture to investigate the astrocyte-derived factors. Therefore, to confirm the experimental results by using the in vitro cell culture-acquired ACM, we cultured astrocytes with TCM from three different TNBC cell lines. Our data identified that in response to tumor cellderived IL-1β and TNF-α, astrocytes not only continued to express TGF-β2, but the expression increased significantly. This also suggests that the interaction of astrocytes and breast cancer cells in brain microenvironment may lead to a positive-feedback loop and thus stronger tumor growth-promoting effects.
Taken together (Fig. 6), this study has shown for the first time that astrocytes produce more TGF-β2 in response to tumor cellderived IL-1β and TNF-α. When exposed to ACM from astrocytes, the TGF-β2 induces TNBC cells to express more ANGPTL4 via the canonical TGF-β pathway. Expression of ANGPTL4 is associated with an increased ability of TNBC cells to seed and grow in the brain. This study aids in the understanding of how TNBC cells gain the abilities to thrive in the brain. However, questions remain, such as those related to the functional role of ANGPTL4 in BM and transcriptional regulation of ANGPTL4. One possible mechanism for ANGPTL4 facilitation of breast cancer brain metastasis is through disrupting vascular endothelial junctions, much like the mechanism for ANGPTL4 facilitation of breast cancer lung metastasis. 21 Another potential mechanism includes ANGPTL4 binding to integrins β1 and β5. This binding hijacks the integrin/ FAK-regulated pathway, thus conferring anoikis resistance in tumors, leading to increased tumor growth. In this model, astrocyte-mediated upregulation of ANGPTL4 in MDA-MB-231 cells allows for colonization in the brain microenvironment. 18 Even so, this study has provided evidence for a novel target of ANGPTL4 to reduce breast cancer BM. Although targeting TGF-β is complex, as it is known to have roles in both inhibiting and promoting tumor growth, this research brings light to other parts of the complex signaling network involved in breast cancer brain metastases that with further research may bring forth more attractive targets.

METHODS
Preparing astrocyte-conditioned medium (ACM) Primary rat neonatal hippocampal astrocytes were isolated, cultured, and identified in Dr. David Harder's laboratory who has more than 20 years of experience on it, which has been described in our previous work. 17 In this study, isolated cells were plated at a density of~3000 cells per mm 2 . Cells were cultured in Dulbecco's Modified Eagle's medium (DMEM, Gibco) with 10% fetal bovine serum (FBS). Culture media was changed three times per week. Confluent monolayers of astrocytes formed within 7-10 days after the initial plating. The primary cultures were maintained by passaging the cells when they were >90% confluent. When cell confluency was ≥70%, the complete medium was aspirated, and the cells were then incubated in serum-free medium for another 24 h. After that, the culture supernatant was collected by centrifugation to remove the cell pellets and then filtered the supernatant through a 0.45-μm filter-this final obtained medium was named astrocyte-conditioned medium (ACM  To evaluate the effect of ACM on the tumor cells, tumor cells were sequentially passaged in the ACM for indicated numbers of passages, allowing 2-3 days of growth between each passage. To evaluate whether ACM is necessary for ACM-induced tumor cell changes, MDA-MB-231 cells that had been passaged in ACM for five passages (P5A) were re-passaged in unconditioned culture media (DMEM) for another five passages (P10AD). For generating tumor cell-conditioned media (TCM), 500,000 cells were plated in 10-cm dishes in each cell line's respective culture medium with 10% FBS, as described above, for 24 h. The complete medium was aspirated, and cells were then incubated in 6 mL of serum-free medium for 24 h. The TCM was then collected, filtered, and used for the experiments. To evaluate the effect of tumor necrosis factor alpha (TNF-α) and interleukin-1 beta (IL-1β) in the TCM on astrocytes, TCM was preincubated at 37°C for 1 h with neutralizing antibodies for TNF-α (R&D Systems) and IL-1β isotype (Abcam) to neutralize the biological activity of the two cytokines.

Microarray analysis
Cell pellets from MDA-MB-231/P5D (n = 3), MDA-MB-231/P5A (n = 3), and MDA-MB-231/P10AD (n = 3) were sent to Arraystar Inc., Rockville, MD for RNA isolation and microarray analysis. RNA was isolated by using TRIzol reagent (Life Technologies). Methodologies related to RNA labeling, array hybridization, and data analysis are available upon request, and are also available directly from Arraystar Inc. The gene-level profiles were analyzed for differentially expressed genes among different treated cells as above by using R-3.5 Software.

In vivo experiment
In vivo experiments were performed as described before. 17 Briefly, 6-8week-old male CB17/SCID mice (CB17/Icr-Prkdc/IcrCrl, Charles Rivers, Wilmington, MA, USA) were used in this study. Mice were inoculated with 1 × 10 5 exponential growth phase tumor cells in a 100-µL volume via left ventricular injection. Ultrasound analysis was done by using a VisualSonics Vevo 770 to visualize the injection needle to ensure left heart ventricle injection of the cells (Additional file 3: Fig. S2A). Post injection, tumor growth was checked by imaging anesthetized mice injected with 200 µl (15 mg/ml) of D-Luciferin (Xenogen Corporation) at indicated time points by using the Lumina IVIS-100 in vivo Imaging System (Xenogen Corporation) (Additional file 3: Fig. S2A). Survival endpoint was determined based on the mouse survival rate ≤30% appearing in any one of the groups. Regions of interest were created and measured as area flux, defined by radiance (photons per second per square centimeter per steradian) according to the manufacturer's calibration (Xenogen Corporation). The MCW Animal Care and Use Committee approved all animal procedures under protocol AUA1022.

Transient siRNA transfection
See supplemental file.

Immunoprecipitation (IP) and immunoblot (IB)
See supplemental file. All blots were derived from the same experiment and were processed in parallel. The full, uncropped images of all blots were included in the supplemental file: Fig. S3.

Transcriptomics analysis
Comparisons of IL-1β and TNF-α expression in different breast cancer subtypes were performed by using publicly available transcriptome data, the Cancer Genome Atlas (TCGA), in UALCAN. 48 Microarray gene expression data from 719 human individuals representing three different molecular subtypes of breast cancer were included in this study. X. Gong et al.